src/shared/dissect-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #if HAVE_VALGRIND_MEMCHECK_H
   4 #include <valgrind/memcheck.h>
   5 #endif
   6
   7 #include <linux/dm-ioctl.h>
   8 #include <linux/loop.h>
   9 #include <sys/mount.h>
  10 #include <sys/prctl.h>
  11 #include <sys/wait.h>
  12 #include <sysexits.h>
  13
  14 #include "sd-device.h"
  15 #include "sd-id128.h"
  16
  17 #include "architecture.h"
  18 #include "ask-password-api.h"
  19 #include "blkid-util.h"
  20 #include "blockdev-util.h"
  21 #include "copy.h"
  22 #include "crypt-util.h"
  23 #include "def.h"
  24 #include "device-nodes.h"
  25 #include "device-util.h"
  26 #include "dissect-image.h"
  27 #include "dm-util.h"
  28 #include "env-file.h"
  29 #include "fd-util.h"
  30 #include "fileio.h"
  31 #include "fs-util.h"
  32 #include "fsck-util.h"
  33 #include "gpt.h"
  34 #include "hexdecoct.h"
  35 #include "hostname-util.h"
  36 #include "id128-util.h"
  37 #include "mkdir.h"
  38 #include "mount-util.h"
  39 #include "mountpoint-util.h"
  40 #include "namespace-util.h"
  41 #include "nulstr-util.h"
  42 #include "os-util.h"
  43 #include "path-util.h"
  44 #include "process-util.h"
  45 #include "raw-clone.h"
  46 #include "signal-util.h"
  47 #include "stat-util.h"
  48 #include "stdio-util.h"
  49 #include "string-table.h"
  50 #include "string-util.h"
  51 #include "strv.h"
  52 #include "tmpfile-util.h"
  53 #include "udev-util.h"
  54 #include "user-util.h"
  55 #include "xattr-util.h"
  56
  57 int probe_filesystem(const char *node, char **ret_fstype) {
  58         /* Try to find device content type and return it in *ret_fstype. If nothing is found,
  59          * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
  60          * different error otherwise. */
  61
  62 #if HAVE_BLKID
  63         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
  64         const char *fstype;
  65         int r;
  66
  67         errno = 0;
  68         b = blkid_new_probe_from_filename(node);
  69         if (!b)
  70                 return errno_or_else(ENOMEM);
  71
  72         blkid_probe_enable_superblocks(b, 1);
  73         blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
  74
  75         errno = 0;
  76         r = blkid_do_safeprobe(b);
  77         if (r == 1) {
  78                 log_debug("No type detected on partition %s", node);
  79                 goto not_found;
  80         }
  81         if (r == -2)
  82                 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
  83                                        "Results ambiguous for partition %s", node);
  84         if (r != 0)
  85                 return errno_or_else(EIO);
  86
  87         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
  88
  89         if (fstype) {
  90                 char *t;
  91
  92                 t = strdup(fstype);
  93                 if (!t)
  94                         return -ENOMEM;
  95
  96                 *ret_fstype = t;
  97                 return 1;
  98         }
  99
 100 not_found:
 101         *ret_fstype = NULL;
 102         return 0;
 103 #else
 104         return -EOPNOTSUPP;
 105 #endif
 106 }
 107
 108 #if HAVE_BLKID
 109 /* Detect RPMB and Boot partitions, which are not listed by blkid.
 110  * See https://github.com/systemd/systemd/issues/5806. */
 111 static bool device_is_mmc_special_partition(sd_device *d) {
 112         const char *sysname;
 113
 114         assert(d);
 115
 116         if (sd_device_get_sysname(d, &sysname) < 0)
 117                 return false;
 118
 119         return startswith(sysname, "mmcblk") &&
 120                 (endswith(sysname, "rpmb") || endswith(sysname, "boot0") || endswith(sysname, "boot1"));
 121 }
 122
 123 static bool device_is_block(sd_device *d) {
 124         const char *ss;
 125
 126         assert(d);
 127
 128         if (sd_device_get_subsystem(d, &ss) < 0)
 129                 return false;
 130
 131         return streq(ss, "block");
 132 }
 133
 134 static int enumerator_for_parent(sd_device *d, sd_device_enumerator **ret) {
 135         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 136         int r;
 137
 138         assert(d);
 139         assert(ret);
 140
 141         r = sd_device_enumerator_new(&e);
 142         if (r < 0)
 143                 return r;
 144
 145         r = sd_device_enumerator_allow_uninitialized(e);
 146         if (r < 0)
 147                 return r;
 148
 149         r = sd_device_enumerator_add_match_parent(e, d);
 150         if (r < 0)
 151                 return r;
 152
 153         *ret = TAKE_PTR(e);
 154         return 0;
 155 }
 156
 157 /* how many times to wait for the device nodes to appear */
 158 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
 159
 160 static int wait_for_partitions_to_appear(
 161                 int fd,
 162                 sd_device *d,
 163                 unsigned num_partitions,
 164                 DissectImageFlags flags,
 165                 sd_device_enumerator **ret_enumerator) {
 166
 167         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 168         sd_device *q;
 169         unsigned n;
 170         int r;
 171
 172         assert(fd >= 0);
 173         assert(d);
 174         assert(ret_enumerator);
 175
 176         r = enumerator_for_parent(d, &e);
 177         if (r < 0)
 178                 return r;
 179
 180         /* Count the partitions enumerated by the kernel */
 181         n = 0;
 182         FOREACH_DEVICE(e, q) {
 183                 if (sd_device_get_devnum(q, NULL) < 0)
 184                         continue;
 185                 if (!device_is_block(q))
 186                         continue;
 187                 if (device_is_mmc_special_partition(q))
 188                         continue;
 189
 190                 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 191                         r = device_wait_for_initialization(q, "block", USEC_INFINITY, NULL);
 192                         if (r < 0)
 193                                 return r;
 194                 }
 195
 196                 n++;
 197         }
 198
 199         if (n == num_partitions + 1) {
 200                 *ret_enumerator = TAKE_PTR(e);
 201                 return 0; /* success! */
 202         }
 203         if (n > num_partitions + 1)
 204                 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
 205                                        "blkid and kernel partition lists do not match.");
 206
 207         /* The kernel has probed fewer partitions than blkid? Maybe the kernel prober is still running or it
 208          * got EBUSY because udev already opened the device. Let's reprobe the device, which is a synchronous
 209          * call that waits until probing is complete. */
 210
 211         for (unsigned j = 0; ; j++) {
 212                 if (j++ > 20)
 213                         return -EBUSY;
 214
 215                 if (ioctl(fd, BLKRRPART, 0) >= 0)
 216                         break;
 217                 r = -errno;
 218                 if (r == -EINVAL) {
 219                         /* If we are running on a block device that has partition scanning off, return an
 220                          * explicit recognizable error about this, so that callers can generate a proper
 221                          * message explaining the situation. */
 222
 223                         r = blockdev_partscan_enabled(fd);
 224                         if (r < 0)
 225                                 return r;
 226                         if (r == 0)
 227                                 return log_debug_errno(EPROTONOSUPPORT,
 228                                                        "Device is a loop device and partition scanning is off!");
 229
 230                         return -EINVAL; /* original error */
 231                 }
 232                 if (r != -EBUSY)
 233                         return r;
 234
 235                 /* If something else has the device open, such as an udev rule, the ioctl will return
 236                  * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
 237                  * and try again.
 238                  *
 239                  * This is really something they should fix in the kernel! */
 240                 (void) usleep(50 * USEC_PER_MSEC);
 241
 242         }
 243
 244         return -EAGAIN; /* no success yet, try again */
 245 }
 246
 247 static int loop_wait_for_partitions_to_appear(
 248                 int fd,
 249                 sd_device *d,
 250                 unsigned num_partitions,
 251                 DissectImageFlags flags,
 252                 sd_device_enumerator **ret_enumerator) {
 253         _cleanup_(sd_device_unrefp) sd_device *device = NULL;
 254         int r;
 255
 256         assert(fd >= 0);
 257         assert(d);
 258         assert(ret_enumerator);
 259
 260         log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions);
 261
 262         if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 263                 r = device_wait_for_initialization(d, "block", USEC_INFINITY, &device);
 264                 if (r < 0)
 265                         return r;
 266         } else
 267                 device = sd_device_ref(d);
 268
 269         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
 270                 r = wait_for_partitions_to_appear(fd, device, num_partitions, flags, ret_enumerator);
 271                 if (r != -EAGAIN)
 272                         return r;
 273         }
 274
 275         return log_debug_errno(SYNTHETIC_ERRNO(ENXIO),
 276                                "Kernel partitions dit not appear within %d attempts",
 277                                N_DEVICE_NODE_LIST_ATTEMPTS);
 278 }
 279
 280 static void check_partition_flags(
 281                 const char *node,
 282                 unsigned long long pflags,
 283                 unsigned long long supported) {
 284
 285         assert(node);
 286
 287         /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
 288         pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
 289
 290         if (pflags == 0)
 291                 return;
 292
 293         /* If there are other bits set, then log about it, to make things discoverable */
 294         for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
 295                 unsigned long long bit = 1ULL << i;
 296                 if (!FLAGS_SET(pflags, bit))
 297                         continue;
 298
 299                 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
 300         }
 301 }
 302
 303 #endif
 304
 305 int dissect_image(
 306                 int fd,
 307                 const void *root_hash,
 308                 size_t root_hash_size,
 309                 const char *verity_data,
 310                 const MountOptions *mount_options,
 311                 DissectImageFlags flags,
 312                 DissectedImage **ret) {
 313
 314 #if HAVE_BLKID
 315         sd_id128_t root_uuid = SD_ID128_NULL, verity_uuid = SD_ID128_NULL;
 316         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 317         bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
 318         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
 319         _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 320         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 321         _cleanup_free_ char *generic_node = NULL;
 322         sd_id128_t generic_uuid = SD_ID128_NULL;
 323         const char *pttype = NULL;
 324         blkid_partlist pl;
 325         int r, generic_nr;
 326         struct stat st;
 327         sd_device *q;
 328         unsigned i;
 329
 330         assert(fd >= 0);
 331         assert(ret);
 332         assert(root_hash || root_hash_size == 0);
 333         assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
 334
 335         /* Probes a disk image, and returns information about what it found in *ret.
 336          *
 337          * Returns -ENOPKG if no suitable partition table or file system could be found.
 338          * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
 339
 340         if (root_hash) {
 341                 /* If a root hash is supplied, then we use the root partition that has a UUID that match the first
 342                  * 128bit of the root hash. And we use the verity partition that has a UUID that match the final
 343                  * 128bit. */
 344
 345                 if (root_hash_size < sizeof(sd_id128_t))
 346                         return -EINVAL;
 347
 348                 memcpy(&root_uuid, root_hash, sizeof(sd_id128_t));
 349                 memcpy(&verity_uuid, (const uint8_t*) root_hash + root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
 350
 351                 if (sd_id128_is_null(root_uuid))
 352                         return -EINVAL;
 353                 if (sd_id128_is_null(verity_uuid))
 354                         return -EINVAL;
 355         }
 356
 357         if (fstat(fd, &st) < 0)
 358                 return -errno;
 359
 360         if (!S_ISBLK(st.st_mode))
 361                 return -ENOTBLK;
 362
 363         b = blkid_new_probe();
 364         if (!b)
 365                 return -ENOMEM;
 366
 367         errno = 0;
 368         r = blkid_probe_set_device(b, fd, 0, 0);
 369         if (r != 0)
 370                 return errno_or_else(ENOMEM);
 371
 372         if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
 373                 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
 374                 blkid_probe_enable_superblocks(b, 1);
 375                 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
 376         }
 377
 378         blkid_probe_enable_partitions(b, 1);
 379         blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
 380
 381         errno = 0;
 382         r = blkid_do_safeprobe(b);
 383         if (IN_SET(r, -2, 1))
 384                 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
 385         if (r != 0)
 386                 return errno_or_else(EIO);
 387
 388         m = new0(DissectedImage, 1);
 389         if (!m)
 390                 return -ENOMEM;
 391
 392         r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
 393         if (r < 0)
 394                 return r;
 395
 396         if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
 397             (flags & DISSECT_IMAGE_REQUIRE_ROOT)) ||
 398             (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
 399                 const char *usage = NULL;
 400
 401                 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
 402                 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
 403                         _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 404                         const char *fstype = NULL, *options = NULL;
 405
 406                         /* OK, we have found a file system, that's our root partition then. */
 407                         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 408
 409                         if (fstype) {
 410                                 t = strdup(fstype);
 411                                 if (!t)
 412                                         return -ENOMEM;
 413                         }
 414
 415                         r = device_path_make_major_minor(st.st_mode, st.st_rdev, &n);
 416                         if (r < 0)
 417                                 return r;
 418
 419                         m->single_file_system = true;
 420                         m->verity = root_hash && verity_data;
 421                         m->can_verity = !!verity_data;
 422
 423                         options = mount_options_from_part(mount_options, 0);
 424                         if (options) {
 425                                 o = strdup(options);
 426                                 if (!o)
 427                                         return -ENOMEM;
 428                         }
 429
 430                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 431                                 .found = true,
 432                                 .rw = !m->verity,
 433                                 .partno = -1,
 434                                 .architecture = _ARCHITECTURE_INVALID,
 435                                 .fstype = TAKE_PTR(t),
 436                                 .node = TAKE_PTR(n),
 437                                 .mount_options = TAKE_PTR(o),
 438                         };
 439
 440                         m->encrypted = streq_ptr(fstype, "crypto_LUKS");
 441
 442                         /* Even on a single partition we need to wait for udev to create the
 443                          * /dev/block/X:Y symlink to /dev/loopZ */
 444                         r = loop_wait_for_partitions_to_appear(fd, d, 0, flags, &e);
 445                         if (r < 0)
 446                                 return r;
 447                         *ret = TAKE_PTR(m);
 448
 449                         return 0;
 450                 }
 451         }
 452
 453         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 454         if (!pttype)
 455                 return -ENOPKG;
 456
 457         is_gpt = streq_ptr(pttype, "gpt");
 458         is_mbr = streq_ptr(pttype, "dos");
 459
 460         if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
 461                 return -ENOPKG;
 462
 463         errno = 0;
 464         pl = blkid_probe_get_partitions(b);
 465         if (!pl)
 466                 return errno_or_else(ENOMEM);
 467
 468         r = loop_wait_for_partitions_to_appear(fd, d, blkid_partlist_numof_partitions(pl), flags, &e);
 469         if (r < 0)
 470                 return r;
 471
 472         FOREACH_DEVICE(e, q) {
 473                 unsigned long long pflags;
 474                 blkid_partition pp;
 475                 const char *node;
 476                 dev_t qn;
 477                 int nr;
 478
 479                 r = sd_device_get_devnum(q, &qn);
 480                 if (r < 0)
 481                         continue;
 482
 483                 if (st.st_rdev == qn)
 484                         continue;
 485
 486                 if (!device_is_block(q))
 487                         continue;
 488
 489                 if (device_is_mmc_special_partition(q))
 490                         continue;
 491
 492                 r = sd_device_get_devname(q, &node);
 493                 if (r < 0)
 494                         continue;
 495
 496                 pp = blkid_partlist_devno_to_partition(pl, qn);
 497                 if (!pp)
 498                         continue;
 499
 500                 pflags = blkid_partition_get_flags(pp);
 501
 502                 nr = blkid_partition_get_partno(pp);
 503                 if (nr < 0)
 504                         continue;
 505
 506                 if (is_gpt) {
 507                         int designator = _PARTITION_DESIGNATOR_INVALID, architecture = _ARCHITECTURE_INVALID;
 508                         const char *stype, *sid, *fstype = NULL;
 509                         sd_id128_t type_id, id;
 510                         bool rw = true;
 511
 512                         sid = blkid_partition_get_uuid(pp);
 513                         if (!sid)
 514                                 continue;
 515                         if (sd_id128_from_string(sid, &id) < 0)
 516                                 continue;
 517
 518                         stype = blkid_partition_get_type_string(pp);
 519                         if (!stype)
 520                                 continue;
 521                         if (sd_id128_from_string(stype, &type_id) < 0)
 522                                 continue;
 523
 524                         if (sd_id128_equal(type_id, GPT_HOME)) {
 525
 526                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 527
 528                                 if (pflags & GPT_FLAG_NO_AUTO)
 529                                         continue;
 530
 531                                 designator = PARTITION_HOME;
 532                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 533                         } else if (sd_id128_equal(type_id, GPT_SRV)) {
 534
 535                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 536
 537                                 if (pflags & GPT_FLAG_NO_AUTO)
 538                                         continue;
 539
 540                                 designator = PARTITION_SRV;
 541                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 542                         } else if (sd_id128_equal(type_id, GPT_ESP)) {
 543
 544                                 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
 545                                  * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
 546                                  * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
 547
 548                                 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
 549                                         continue;
 550
 551                                 designator = PARTITION_ESP;
 552                                 fstype = "vfat";
 553
 554                         } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
 555
 556                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 557
 558                                 if (pflags & GPT_FLAG_NO_AUTO)
 559                                         continue;
 560
 561                                 designator = PARTITION_XBOOTLDR;
 562                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 563                         }
 564 #ifdef GPT_ROOT_NATIVE
 565                         else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
 566
 567                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 568
 569                                 if (pflags & GPT_FLAG_NO_AUTO)
 570                                         continue;
 571
 572                                 /* If a root ID is specified, ignore everything but the root id */
 573                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 574                                         continue;
 575
 576                                 designator = PARTITION_ROOT;
 577                                 architecture = native_architecture();
 578                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 579                         } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
 580
 581                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 582
 583                                 if (pflags & GPT_FLAG_NO_AUTO)
 584                                         continue;
 585
 586                                 m->can_verity = true;
 587
 588                                 /* Ignore verity unless a root hash is specified */
 589                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 590                                         continue;
 591
 592                                 designator = PARTITION_ROOT_VERITY;
 593                                 fstype = "DM_verity_hash";
 594                                 architecture = native_architecture();
 595                                 rw = false;
 596                         }
 597 #endif
 598 #ifdef GPT_ROOT_SECONDARY
 599                         else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
 600
 601                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 602
 603                                 if (pflags & GPT_FLAG_NO_AUTO)
 604                                         continue;
 605
 606                                 /* If a root ID is specified, ignore everything but the root id */
 607                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 608                                         continue;
 609
 610                                 designator = PARTITION_ROOT_SECONDARY;
 611                                 architecture = SECONDARY_ARCHITECTURE;
 612                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 613                         } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
 614
 615                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 616
 617                                 if (pflags & GPT_FLAG_NO_AUTO)
 618                                         continue;
 619
 620                                 m->can_verity = true;
 621
 622                                 /* Ignore verity unless root has is specified */
 623                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 624                                         continue;
 625
 626                                 designator = PARTITION_ROOT_SECONDARY_VERITY;
 627                                 fstype = "DM_verity_hash";
 628                                 architecture = SECONDARY_ARCHITECTURE;
 629                                 rw = false;
 630                         }
 631 #endif
 632                         else if (sd_id128_equal(type_id, GPT_SWAP)) {
 633
 634                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
 635
 636                                 if (pflags & GPT_FLAG_NO_AUTO)
 637                                         continue;
 638
 639                                 designator = PARTITION_SWAP;
 640                                 fstype = "swap";
 641                         } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
 642
 643                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 644
 645                                 if (pflags & GPT_FLAG_NO_AUTO)
 646                                         continue;
 647
 648                                 if (generic_node)
 649                                         multiple_generic = true;
 650                                 else {
 651                                         generic_nr = nr;
 652                                         generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
 653                                         generic_uuid = id;
 654                                         generic_node = strdup(node);
 655                                         if (!generic_node)
 656                                                 return -ENOMEM;
 657                                 }
 658
 659                         } else if (sd_id128_equal(type_id, GPT_TMP)) {
 660
 661                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 662
 663                                 if (pflags & GPT_FLAG_NO_AUTO)
 664                                         continue;
 665
 666                                 designator = PARTITION_TMP;
 667                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 668
 669                         } else if (sd_id128_equal(type_id, GPT_VAR)) {
 670
 671                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 672
 673                                 if (pflags & GPT_FLAG_NO_AUTO)
 674                                         continue;
 675
 676                                 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
 677                                         sd_id128_t var_uuid;
 678
 679                                         /* For /var we insist that the uuid of the partition matches the
 680                                          * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
 681                                          * ID. Why? Unlike the other partitions /var is inherently
 682                                          * installation specific, hence we need to be careful not to mount it
 683                                          * in the wrong installation. By hashing the partition UUID from
 684                                          * /etc/machine-id we can securely bind the partition to the
 685                                          * installation. */
 686
 687                                         r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
 688                                         if (r < 0)
 689                                                 return r;
 690
 691                                         if (!sd_id128_equal(var_uuid, id)) {
 692                                                 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
 693                                                 continue;
 694                                         }
 695                                 }
 696
 697                                 designator = PARTITION_VAR;
 698                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 699                         }
 700
 701                         if (designator != _PARTITION_DESIGNATOR_INVALID) {
 702                                 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 703                                 const char *options = NULL;
 704
 705                                 /* First one wins */
 706                                 if (m->partitions[designator].found)
 707                                         continue;
 708
 709                                 if (fstype) {
 710                                         t = strdup(fstype);
 711                                         if (!t)
 712                                                 return -ENOMEM;
 713                                 }
 714
 715                                 n = strdup(node);
 716                                 if (!n)
 717                                         return -ENOMEM;
 718
 719                                 options = mount_options_from_part(mount_options, nr);
 720                                 if (options) {
 721                                         o = strdup(options);
 722                                         if (!o)
 723                                                 return -ENOMEM;
 724                                 }
 725
 726                                 m->partitions[designator] = (DissectedPartition) {
 727                                         .found = true,
 728                                         .partno = nr,
 729                                         .rw = rw,
 730                                         .architecture = architecture,
 731                                         .node = TAKE_PTR(n),
 732                                         .fstype = TAKE_PTR(t),
 733                                         .uuid = id,
 734                                         .mount_options = TAKE_PTR(o),
 735                                 };
 736                         }
 737
 738                 } else if (is_mbr) {
 739
 740                         switch (blkid_partition_get_type(pp)) {
 741
 742                         case 0x83: /* Linux partition */
 743
 744                                 if (pflags != 0x80) /* Bootable flag */
 745                                         continue;
 746
 747                                 if (generic_node)
 748                                         multiple_generic = true;
 749                                 else {
 750                                         generic_nr = nr;
 751                                         generic_rw = true;
 752                                         generic_node = strdup(node);
 753                                         if (!generic_node)
 754                                                 return -ENOMEM;
 755                                 }
 756
 757                                 break;
 758
 759                         case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
 760                                 _cleanup_free_ char *n = NULL, *o = NULL;
 761                                 sd_id128_t id = SD_ID128_NULL;
 762                                 const char *sid, *options = NULL;
 763
 764                                 /* First one wins */
 765                                 if (m->partitions[PARTITION_XBOOTLDR].found)
 766                                         continue;
 767
 768                                 sid = blkid_partition_get_uuid(pp);
 769                                 if (sid)
 770                                         (void) sd_id128_from_string(sid, &id);
 771
 772                                 n = strdup(node);
 773                                 if (!n)
 774                                         return -ENOMEM;
 775
 776                                 options = mount_options_from_part(mount_options, nr);
 777                                 if (options) {
 778                                         o = strdup(options);
 779                                         if (!o)
 780                                                 return -ENOMEM;
 781                                 }
 782
 783                                 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
 784                                         .found = true,
 785                                         .partno = nr,
 786                                         .rw = true,
 787                                         .architecture = _ARCHITECTURE_INVALID,
 788                                         .node = TAKE_PTR(n),
 789                                         .uuid = id,
 790                                         .mount_options = TAKE_PTR(o),
 791                                 };
 792
 793                                 break;
 794                         }}
 795                 }
 796         }
 797
 798         if (!m->partitions[PARTITION_ROOT].found) {
 799                 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
 800                  * either, then check if there's a single generic one, and use that. */
 801
 802                 if (m->partitions[PARTITION_ROOT_VERITY].found)
 803                         return -EADDRNOTAVAIL;
 804
 805                 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
 806                         m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
 807                         zero(m->partitions[PARTITION_ROOT_SECONDARY]);
 808
 809                         m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
 810                         zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
 811
 812                 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
 813                         _cleanup_free_ char *o = NULL;
 814                         const char *options = NULL;
 815
 816                         /* If the root hash was set, then we won't fall back to a generic node, because the
 817                          * root hash decides. */
 818                         if (root_hash)
 819                                 return -EADDRNOTAVAIL;
 820
 821                         /* If we didn't find a generic node, then we can't fix this up either */
 822                         if (!generic_node)
 823                                 return -ENXIO;
 824
 825                         /* If we didn't find a properly marked root partition, but we did find a single suitable
 826                          * generic Linux partition, then use this as root partition, if the caller asked for it. */
 827                         if (multiple_generic)
 828                                 return -ENOTUNIQ;
 829
 830                         options = mount_options_from_part(mount_options, generic_nr);
 831                         if (options) {
 832                                 o = strdup(options);
 833                                 if (!o)
 834                                         return -ENOMEM;
 835                         }
 836
 837                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 838                                 .found = true,
 839                                 .rw = generic_rw,
 840                                 .partno = generic_nr,
 841                                 .architecture = _ARCHITECTURE_INVALID,
 842                                 .node = TAKE_PTR(generic_node),
 843                                 .uuid = generic_uuid,
 844                                 .mount_options = TAKE_PTR(o),
 845                         };
 846                 }
 847         }
 848
 849         if (root_hash) {
 850                 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
 851                         return -EADDRNOTAVAIL;
 852
 853                 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
 854                  * (which would be weird, after all the root hash should only be assigned to one pair of
 855                  * partitions... */
 856                 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
 857                 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
 858
 859                 /* If we found a verity setup, then the root partition is necessarily read-only. */
 860                 m->partitions[PARTITION_ROOT].rw = false;
 861
 862                 m->verity = true;
 863         }
 864
 865         blkid_free_probe(b);
 866         b = NULL;
 867
 868         /* Fill in file system types if we don't know them yet. */
 869         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 870                 DissectedPartition *p = m->partitions + i;
 871
 872                 if (!p->found)
 873                         continue;
 874
 875                 if (!p->fstype && p->node) {
 876                         r = probe_filesystem(p->node, &p->fstype);
 877                         if (r < 0 && r != -EUCLEAN)
 878                                 return r;
 879                 }
 880
 881                 if (streq_ptr(p->fstype, "crypto_LUKS"))
 882                         m->encrypted = true;
 883
 884                 if (p->fstype && fstype_is_ro(p->fstype))
 885                         p->rw = false;
 886         }
 887
 888         *ret = TAKE_PTR(m);
 889
 890         return 0;
 891 #else
 892         return -EOPNOTSUPP;
 893 #endif
 894 }
 895
 896 DissectedImage* dissected_image_unref(DissectedImage *m) {
 897         unsigned i;
 898
 899         if (!m)
 900                 return NULL;
 901
 902         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 903                 free(m->partitions[i].fstype);
 904                 free(m->partitions[i].node);
 905                 free(m->partitions[i].decrypted_fstype);
 906                 free(m->partitions[i].decrypted_node);
 907                 free(m->partitions[i].mount_options);
 908         }
 909
 910         free(m->hostname);
 911         strv_free(m->machine_info);
 912         strv_free(m->os_release);
 913
 914         return mfree(m);
 915 }
 916
 917 static int is_loop_device(const char *path) {
 918         char s[SYS_BLOCK_PATH_MAX("/../loop/")];
 919         struct stat st;
 920
 921         assert(path);
 922
 923         if (stat(path, &st) < 0)
 924                 return -errno;
 925
 926         if (!S_ISBLK(st.st_mode))
 927                 return -ENOTBLK;
 928
 929         xsprintf_sys_block_path(s, "/loop/", st.st_dev);
 930         if (access(s, F_OK) < 0) {
 931                 if (errno != ENOENT)
 932                         return -errno;
 933
 934                 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
 935                 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
 936                 if (access(s, F_OK) < 0)
 937                         return errno == ENOENT ? false : -errno;
 938         }
 939
 940         return true;
 941 }
 942
 943 static int run_fsck(const char *node, const char *fstype) {
 944         int r, exit_status;
 945         pid_t pid;
 946
 947         assert(node);
 948         assert(fstype);
 949
 950         r = fsck_exists(fstype);
 951         if (r < 0) {
 952                 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
 953                 return 0;
 954         }
 955         if (r == 0) {
 956                 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
 957                 return 0;
 958         }
 959
 960         r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
 961         if (r < 0)
 962                 return log_debug_errno(r, "Failed to fork off fsck: %m");
 963         if (r == 0) {
 964                 /* Child */
 965                 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
 966                 log_debug_errno(errno, "Failed to execl() fsck: %m");
 967                 _exit(FSCK_OPERATIONAL_ERROR);
 968         }
 969
 970         exit_status = wait_for_terminate_and_check("fsck", pid, 0);
 971         if (exit_status < 0)
 972                 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
 973
 974         if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
 975                 log_debug("fsck failed with exit status %i.", exit_status);
 976
 977                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 978                         return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
 979
 980                 log_debug("Ignoring fsck error.");
 981         }
 982
 983         return 0;
 984 }
 985
 986 static int mount_partition(
 987                 DissectedPartition *m,
 988                 const char *where,
 989                 const char *directory,
 990                 uid_t uid_shift,
 991                 DissectImageFlags flags) {
 992
 993         _cleanup_free_ char *chased = NULL, *options = NULL;
 994         const char *p, *node, *fstype;
 995         bool rw;
 996         int r;
 997
 998         assert(m);
 999         assert(where);
1000
1001         node = m->decrypted_node ?: m->node;
1002         fstype = m->decrypted_fstype ?: m->fstype;
1003
1004         if (!m->found || !node || !fstype)
1005                 return 0;
1006
1007         /* We are looking at an encrypted partition? This either means stacked encryption, or the caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this case. */
1008         if (streq_ptr(fstype, "crypto_LUKS"))
1009                 return -EUNATCH;
1010
1011         rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
1012
1013         if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1014                 r = run_fsck(node, fstype);
1015                 if (r < 0)
1016                         return r;
1017         }
1018
1019         if (directory) {
1020                 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1021                 if (r < 0)
1022                         return r;
1023
1024                 p = chased;
1025         } else
1026                 p = where;
1027
1028         /* If requested, turn on discard support. */
1029         if (fstype_can_discard(fstype) &&
1030             ((flags & DISSECT_IMAGE_DISCARD) ||
1031              ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node)))) {
1032                 options = strdup("discard");
1033                 if (!options)
1034                         return -ENOMEM;
1035         }
1036
1037         if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
1038                 _cleanup_free_ char *uid_option = NULL;
1039
1040                 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1041                         return -ENOMEM;
1042
1043                 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1044                         return -ENOMEM;
1045         }
1046
1047         if (!isempty(m->mount_options))
1048                 if (!strextend_with_separator(&options, ",", m->mount_options, NULL))
1049                         return -ENOMEM;
1050
1051         if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1052                 r = mkdir_p(p, 0755);
1053                 if (r < 0)
1054                         return r;
1055         }
1056
1057         r = mount_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1058         if (r < 0)
1059                 return r;
1060
1061         return 1;
1062 }
1063
1064 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1065         int r, boot_mounted;
1066
1067         assert(m);
1068         assert(where);
1069
1070         /* Returns:
1071          *
1072          *  -ENXIO        → No root partition found
1073          *  -EMEDIUMTYPE  → DISSECT_IMAGE_VALIDATE_OS set but no os-release file found
1074          *  -EUNATCH      → Encrypted partition found for which no dm-crypt was set up yet
1075          *  -EUCLEAN      → fsck for file system failed
1076          *  -EBUSY        → File system already mounted/used elsewhere (kernel)
1077          */
1078
1079         if (!m->partitions[PARTITION_ROOT].found)
1080                 return -ENXIO;
1081
1082         if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1083                 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1084                 if (r < 0)
1085                         return r;
1086
1087                 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1088                         r = path_is_os_tree(where);
1089                         if (r < 0)
1090                                 return r;
1091                         if (r == 0)
1092                                 return -EMEDIUMTYPE;
1093                 }
1094         }
1095
1096         if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1097                 return 0;
1098
1099         /* Mask DISSECT_IMAGE_MKDIR for all subdirs: the idea is that only the top-level mount point is
1100          * created if needed, but the image itself not modified. */
1101         flags &= ~DISSECT_IMAGE_MKDIR;
1102
1103         r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1104         if (r < 0)
1105                 return r;
1106
1107         r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1108         if (r < 0)
1109                 return r;
1110
1111         r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1112         if (r < 0)
1113                 return r;
1114
1115         r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1116         if (r < 0)
1117                 return r;
1118
1119         boot_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1120         if (boot_mounted < 0)
1121                 return boot_mounted;
1122
1123         if (m->partitions[PARTITION_ESP].found) {
1124                 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1125                  * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1126
1127                 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1128                 if (r >= 0) {
1129                         r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1130                         if (r < 0)
1131                                 return r;
1132
1133                 } else if (boot_mounted <= 0) {
1134                         _cleanup_free_ char *p = NULL;
1135
1136                         r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1137                         if (r >= 0 && dir_is_empty(p) > 0) {
1138                                 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1139                                 if (r < 0)
1140                                         return r;
1141                         }
1142                 }
1143         }
1144
1145         return 0;
1146 }
1147
1148 int dissected_image_mount_and_warn(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1149         int r;
1150
1151         assert(m);
1152         assert(where);
1153
1154         r = dissected_image_mount(m, where, uid_shift, flags);
1155         if (r == -ENXIO)
1156                 return log_error_errno(r, "Not root file system found in image.");
1157         if (r == -EMEDIUMTYPE)
1158                 return log_error_errno(r, "No suitable os-release file in image found.");
1159         if (r == -EUNATCH)
1160                 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1161         if (r == -EUCLEAN)
1162                 return log_error_errno(r, "File system check on image failed.");
1163         if (r == -EBUSY)
1164                 return log_error_errno(r, "File system already mounted elsewhere.");
1165         if (r < 0)
1166                 return log_error_errno(r, "Failed to mount image: %m");
1167
1168         return r;
1169 }
1170
1171 #if HAVE_LIBCRYPTSETUP
1172 typedef struct DecryptedPartition {
1173         struct crypt_device *device;
1174         char *name;
1175         bool relinquished;
1176 } DecryptedPartition;
1177
1178 struct DecryptedImage {
1179         DecryptedPartition *decrypted;
1180         size_t n_decrypted;
1181         size_t n_allocated;
1182 };
1183 #endif
1184
1185 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1186 #if HAVE_LIBCRYPTSETUP
1187         size_t i;
1188         int r;
1189
1190         if (!d)
1191                 return NULL;
1192
1193         for (i = 0; i < d->n_decrypted; i++) {
1194                 DecryptedPartition *p = d->decrypted + i;
1195
1196                 if (p->device && p->name && !p->relinquished) {
1197                         r = crypt_deactivate(p->device, p->name);
1198                         if (r < 0)
1199                                 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1200                 }
1201
1202                 if (p->device)
1203                         crypt_free(p->device);
1204                 free(p->name);
1205         }
1206
1207         free(d);
1208 #endif
1209         return NULL;
1210 }
1211
1212 #if HAVE_LIBCRYPTSETUP
1213
1214 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1215         _cleanup_free_ char *name = NULL, *node = NULL;
1216         const char *base;
1217
1218         assert(original_node);
1219         assert(suffix);
1220         assert(ret_name);
1221         assert(ret_node);
1222
1223         base = strrchr(original_node, '/');
1224         if (!base)
1225                 base = original_node;
1226         else
1227                 base++;
1228         if (isempty(base))
1229                 return -EINVAL;
1230
1231         name = strjoin(base, suffix);
1232         if (!name)
1233                 return -ENOMEM;
1234         if (!filename_is_valid(name))
1235                 return -EINVAL;
1236
1237         node = path_join(crypt_get_dir(), name);
1238         if (!node)
1239                 return -ENOMEM;
1240
1241         *ret_name = TAKE_PTR(name);
1242         *ret_node = TAKE_PTR(node);
1243
1244         return 0;
1245 }
1246
1247 static int decrypt_partition(
1248                 DissectedPartition *m,
1249                 const char *passphrase,
1250                 DissectImageFlags flags,
1251                 DecryptedImage *d) {
1252
1253         _cleanup_free_ char *node = NULL, *name = NULL;
1254         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1255         int r;
1256
1257         assert(m);
1258         assert(d);
1259
1260         if (!m->found || !m->node || !m->fstype)
1261                 return 0;
1262
1263         if (!streq(m->fstype, "crypto_LUKS"))
1264                 return 0;
1265
1266         if (!passphrase)
1267                 return -ENOKEY;
1268
1269         r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1270         if (r < 0)
1271                 return r;
1272
1273         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1274                 return -ENOMEM;
1275
1276         r = crypt_init(&cd, m->node);
1277         if (r < 0)
1278                 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1279
1280         cryptsetup_enable_logging(cd);
1281
1282         r = crypt_load(cd, CRYPT_LUKS, NULL);
1283         if (r < 0)
1284                 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1285
1286         r = crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1287                                          ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1288                                          ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1289         if (r < 0) {
1290                 log_debug_errno(r, "Failed to activate LUKS device: %m");
1291                 return r == -EPERM ? -EKEYREJECTED : r;
1292         }
1293
1294         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1295         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1296         d->n_decrypted++;
1297
1298         m->decrypted_node = TAKE_PTR(node);
1299
1300         return 0;
1301 }
1302
1303 static int verity_can_reuse(const void *root_hash, size_t root_hash_size, bool has_sig, const char *name, struct crypt_device **ret_cd) {
1304         /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1305         _cleanup_free_ char *root_hash_existing = NULL;
1306         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1307         struct crypt_params_verity crypt_params = {};
1308         size_t root_hash_existing_size = root_hash_size;
1309         int r;
1310
1311         assert(ret_cd);
1312
1313         r = crypt_init_by_name(&cd, name);
1314         if (r < 0)
1315                 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1316         r = crypt_get_verity_info(cd, &crypt_params);
1317         if (r < 0)
1318                 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1319         root_hash_existing = malloc0(root_hash_size);
1320         if (!root_hash_existing)
1321                 return -ENOMEM;
1322         r = crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1323         if (r < 0)
1324                 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1325         if (root_hash_size != root_hash_existing_size || memcmp(root_hash_existing, root_hash, root_hash_size) != 0)
1326                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1327 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1328         /* Ensure that, if signatures are supported, we only reuse the device if the previous mount
1329          * used the same settings, so that a previous unsigned mount will not be reused if the user
1330          * asks to use signing for the new one, and viceversa. */
1331         if (has_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1332                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1333 #endif
1334
1335         *ret_cd = TAKE_PTR(cd);
1336         return 0;
1337 }
1338
1339 static inline void dm_deferred_remove_clean(char *name) {
1340         if (!name)
1341                 return;
1342         (void) crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1343         free(name);
1344 }
1345 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1346
1347 static int verity_partition(
1348                 DissectedPartition *m,
1349                 DissectedPartition *v,
1350                 const void *root_hash,
1351                 size_t root_hash_size,
1352                 const char *verity_data,
1353                 const char *root_hash_sig_path,
1354                 const void *root_hash_sig,
1355                 size_t root_hash_sig_size,
1356                 DissectImageFlags flags,
1357                 DecryptedImage *d) {
1358
1359         _cleanup_free_ char *node = NULL, *name = NULL, *hash_sig_from_file = NULL;
1360         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1361         _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1362         int r;
1363
1364         assert(m);
1365         assert(v || verity_data);
1366
1367         if (!root_hash)
1368                 return 0;
1369
1370         if (!m->found || !m->node || !m->fstype)
1371                 return 0;
1372         if (!verity_data) {
1373                 if (!v->found || !v->node || !v->fstype)
1374                         return 0;
1375
1376                 if (!streq(v->fstype, "DM_verity_hash"))
1377                         return 0;
1378         }
1379
1380         if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1381                 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1382                 _cleanup_free_ char *root_hash_encoded = NULL;
1383                 root_hash_encoded = hexmem(root_hash, root_hash_size);
1384                 if (!root_hash_encoded)
1385                         return -ENOMEM;
1386                 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1387         } else
1388                 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1389         if (r < 0)
1390                 return r;
1391
1392         if (!root_hash_sig && root_hash_sig_path) {
1393                 r = read_full_file_full(AT_FDCWD, root_hash_sig_path, 0, &hash_sig_from_file, &root_hash_sig_size);
1394                 if (r < 0)
1395                         return r;
1396         }
1397
1398         r = crypt_init(&cd, verity_data ?: v->node);
1399         if (r < 0)
1400                 return r;
1401
1402         cryptsetup_enable_logging(cd);
1403
1404         r = crypt_load(cd, CRYPT_VERITY, NULL);
1405         if (r < 0)
1406                 return r;
1407
1408         r = crypt_set_data_device(cd, m->node);
1409         if (r < 0)
1410                 return r;
1411
1412         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1413                 return -ENOMEM;
1414
1415         /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1416          * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1417          * retry a few times before giving up. */
1418         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1419                 if (root_hash_sig || hash_sig_from_file) {
1420 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1421                         r = crypt_activate_by_signed_key(cd, name, root_hash, root_hash_size, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, CRYPT_ACTIVATE_READONLY);
1422 #else
1423                         r = log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "activation of verity device with signature requested, but not supported by cryptsetup due to missing crypt_activate_by_signed_key()");
1424 #endif
1425                 } else
1426                         r = crypt_activate_by_volume_key(cd, name, root_hash, root_hash_size, CRYPT_ACTIVATE_READONLY);
1427                 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1428                  * There's no way to distinguish this situation from a genuine error due to invalid
1429                  * parameters, so immediately fall back to activating the device with a unique name.
1430                  * Improvements in libcrypsetup can ensure this never happens: https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1431                 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1432                         return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1433                 if (!IN_SET(r,
1434                             0, /* Success */
1435                             -EEXIST, /* Volume is already open and ready to be used */
1436                             -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1437                             -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
1438                         return r;
1439                 if (IN_SET(r, -EEXIST, -EBUSY)) {
1440                         struct crypt_device *existing_cd = NULL;
1441
1442                         if (!restore_deferred_remove){
1443                                 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1444                                 r = dm_deferred_remove_cancel(name);
1445                                 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1446                                 if (r < 0 && r != -ENXIO)
1447                                         return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
1448                                 if (r == 0) {
1449                                         restore_deferred_remove = strdup(name);
1450                                         if (!restore_deferred_remove)
1451                                                 return -ENOMEM;
1452                                 }
1453                         }
1454
1455                         r = verity_can_reuse(root_hash, root_hash_size, !!root_hash_sig || !!hash_sig_from_file, name, &existing_cd);
1456                         /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1457                         if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1458                                 return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1459                         if (!IN_SET(r, 0, -ENODEV, -ENOENT, -EBUSY))
1460                                 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1461                         if (r == 0) {
1462                                 /* devmapper might say that the device exists, but the devlink might not yet have been
1463                                  * created. Check and wait for the udev event in that case. */
1464                                 r = device_wait_for_devlink(node, "block", 100 * USEC_PER_MSEC, NULL);
1465                                 /* Fallback to activation with a unique device if it's taking too long */
1466                                 if (r == -ETIMEDOUT)
1467                                         break;
1468                                 if (r < 0)
1469                                         return r;
1470
1471                                 if (cd)
1472                                         crypt_free(cd);
1473                                 cd = existing_cd;
1474                         }
1475                 }
1476                 if (r == 0)
1477                         break;
1478
1479                 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
1480                 (void) usleep(2 * USEC_PER_MSEC);
1481         }
1482
1483         /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1484          * Fall back to activating it with a unique device name. */
1485         if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1486                 return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1487
1488         /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1489         restore_deferred_remove = mfree(restore_deferred_remove);
1490
1491         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1492         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1493         d->n_decrypted++;
1494
1495         m->decrypted_node = TAKE_PTR(node);
1496
1497         return 0;
1498 }
1499 #endif
1500
1501 int dissected_image_decrypt(
1502                 DissectedImage *m,
1503                 const char *passphrase,
1504                 const void *root_hash,
1505                 size_t root_hash_size,
1506                 const char *verity_data,
1507                 const char *root_hash_sig_path,
1508                 const void *root_hash_sig,
1509                 size_t root_hash_sig_size,
1510                 DissectImageFlags flags,
1511                 DecryptedImage **ret) {
1512
1513 #if HAVE_LIBCRYPTSETUP
1514         _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1515         unsigned i;
1516         int r;
1517 #endif
1518
1519         assert(m);
1520         assert(root_hash || root_hash_size == 0);
1521
1522         /* Returns:
1523          *
1524          *      = 0           → There was nothing to decrypt
1525          *      > 0           → Decrypted successfully
1526          *      -ENOKEY       → There's something to decrypt but no key was supplied
1527          *      -EKEYREJECTED → Passed key was not correct
1528          */
1529
1530         if (root_hash && root_hash_size < sizeof(sd_id128_t))
1531                 return -EINVAL;
1532
1533         if (!m->encrypted && !m->verity) {
1534                 *ret = NULL;
1535                 return 0;
1536         }
1537
1538 #if HAVE_LIBCRYPTSETUP
1539         d = new0(DecryptedImage, 1);
1540         if (!d)
1541                 return -ENOMEM;
1542
1543         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1544                 DissectedPartition *p = m->partitions + i;
1545                 int k;
1546
1547                 if (!p->found)
1548                         continue;
1549
1550                 r = decrypt_partition(p, passphrase, flags, d);
1551                 if (r < 0)
1552                         return r;
1553
1554                 k = PARTITION_VERITY_OF(i);
1555                 if (k >= 0) {
1556                         r = verity_partition(p, m->partitions + k, root_hash, root_hash_size, verity_data, root_hash_sig_path, root_hash_sig, root_hash_sig_size, flags | DISSECT_IMAGE_VERITY_SHARE, d);
1557                         if (r < 0)
1558                                 return r;
1559                 }
1560
1561                 if (!p->decrypted_fstype && p->decrypted_node) {
1562                         r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1563                         if (r < 0 && r != -EUCLEAN)
1564                                 return r;
1565                 }
1566         }
1567
1568         *ret = TAKE_PTR(d);
1569
1570         return 1;
1571 #else
1572         return -EOPNOTSUPP;
1573 #endif
1574 }
1575
1576 int dissected_image_decrypt_interactively(
1577                 DissectedImage *m,
1578                 const char *passphrase,
1579                 const void *root_hash,
1580                 size_t root_hash_size,
1581                 const char *verity_data,
1582                 const char *root_hash_sig_path,
1583                 const void *root_hash_sig,
1584                 size_t root_hash_sig_size,
1585                 DissectImageFlags flags,
1586                 DecryptedImage **ret) {
1587
1588         _cleanup_strv_free_erase_ char **z = NULL;
1589         int n = 3, r;
1590
1591         if (passphrase)
1592                 n--;
1593
1594         for (;;) {
1595                 r = dissected_image_decrypt(m, passphrase, root_hash, root_hash_size, verity_data, root_hash_sig_path, root_hash_sig, root_hash_sig_size, flags, ret);
1596                 if (r >= 0)
1597                         return r;
1598                 if (r == -EKEYREJECTED)
1599                         log_error_errno(r, "Incorrect passphrase, try again!");
1600                 else if (r != -ENOKEY)
1601                         return log_error_errno(r, "Failed to decrypt image: %m");
1602
1603                 if (--n < 0)
1604                         return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1605                                                "Too many retries.");
1606
1607                 z = strv_free(z);
1608
1609                 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1610                 if (r < 0)
1611                         return log_error_errno(r, "Failed to query for passphrase: %m");
1612
1613                 passphrase = z[0];
1614         }
1615 }
1616
1617 int decrypted_image_relinquish(DecryptedImage *d) {
1618
1619 #if HAVE_LIBCRYPTSETUP
1620         size_t i;
1621         int r;
1622 #endif
1623
1624         assert(d);
1625
1626         /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1627          * that we don't clean it up ourselves either anymore */
1628
1629 #if HAVE_LIBCRYPTSETUP
1630         for (i = 0; i < d->n_decrypted; i++) {
1631                 DecryptedPartition *p = d->decrypted + i;
1632
1633                 if (p->relinquished)
1634                         continue;
1635
1636                 r = crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
1637                 if (r < 0)
1638                         return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1639
1640                 p->relinquished = true;
1641         }
1642 #endif
1643
1644         return 0;
1645 }
1646
1647 int verity_metadata_load(
1648                 const char *image,
1649                 const char *root_hash_path,
1650                 void **ret_roothash,
1651                 size_t *ret_roothash_size,
1652                 char **ret_verity_data,
1653                 char **ret_roothashsig) {
1654
1655         _cleanup_free_ char *verity_filename = NULL, *roothashsig_filename = NULL;
1656         _cleanup_free_ void *roothash_decoded = NULL;
1657         size_t roothash_decoded_size = 0;
1658         int r;
1659
1660         assert(image);
1661
1662         if (is_device_path(image)) {
1663                 /* If we are asked to load the root hash for a device node, exit early */
1664                 if (ret_roothash)
1665                         *ret_roothash = NULL;
1666                 if (ret_roothash_size)
1667                         *ret_roothash_size = 0;
1668                 if (ret_verity_data)
1669                         *ret_verity_data = NULL;
1670                 if (ret_roothashsig)
1671                         *ret_roothashsig = NULL;
1672                 return 0;
1673         }
1674
1675         if (ret_verity_data) {
1676                 char *e;
1677
1678                 verity_filename = new(char, strlen(image) + STRLEN(".verity") + 1);
1679                 if (!verity_filename)
1680                         return -ENOMEM;
1681                 strcpy(verity_filename, image);
1682                 e = endswith(verity_filename, ".raw");
1683                 if (e)
1684                         strcpy(e, ".verity");
1685                 else
1686                         strcat(verity_filename, ".verity");
1687
1688                 r = access(verity_filename, F_OK);
1689                 if (r < 0) {
1690                         if (errno != ENOENT)
1691                                 return -errno;
1692                         verity_filename = mfree(verity_filename);
1693                 }
1694         }
1695
1696         if (ret_roothashsig) {
1697                 char *e;
1698
1699                 /* Follow naming convention recommended by the relevant RFC:
1700                  * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
1701                 roothashsig_filename = new(char, strlen(image) + STRLEN(".roothash.p7s") + 1);
1702                 if (!roothashsig_filename)
1703                         return -ENOMEM;
1704                 strcpy(roothashsig_filename, image);
1705                 e = endswith(roothashsig_filename, ".raw");
1706                 if (e)
1707                         strcpy(e, ".roothash.p7s");
1708                 else
1709                         strcat(roothashsig_filename, ".roothash.p7s");
1710
1711                 r = access(roothashsig_filename, R_OK);
1712                 if (r < 0) {
1713                         if (errno != ENOENT)
1714                                 return -errno;
1715                         roothashsig_filename = mfree(roothashsig_filename);
1716                 }
1717         }
1718
1719         if (ret_roothash) {
1720                 _cleanup_free_ char *text = NULL;
1721                 assert(ret_roothash_size);
1722
1723                 if (root_hash_path) {
1724                         /* We have the path to a roothash to load and decode, eg: RootHash=/foo/bar.roothash */
1725                         r = read_one_line_file(root_hash_path, &text);
1726                         if (r < 0)
1727                                 return r;
1728                 } else {
1729                         r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1730                         if (r < 0) {
1731                                 char *fn, *e, *n;
1732
1733                                 if (!IN_SET(r, -ENODATA, -EOPNOTSUPP, -ENOENT))
1734                                         return r;
1735
1736                                 fn = newa(char, strlen(image) + STRLEN(".roothash") + 1);
1737                                 n = stpcpy(fn, image);
1738                                 e = endswith(fn, ".raw");
1739                                 if (e)
1740                                         n = e;
1741
1742                                 strcpy(n, ".roothash");
1743
1744                                 r = read_one_line_file(fn, &text);
1745                                 if (r < 0 && r != -ENOENT)
1746                                         return r;
1747                         }
1748                 }
1749
1750                 if (text) {
1751                         r = unhexmem(text, strlen(text), &roothash_decoded, &roothash_decoded_size);
1752                         if (r < 0)
1753                                 return r;
1754                         if (roothash_decoded_size < sizeof(sd_id128_t))
1755                                 return -EINVAL;
1756                 }
1757         }
1758
1759         if (ret_roothash) {
1760                 *ret_roothash = TAKE_PTR(roothash_decoded);
1761                 *ret_roothash_size = roothash_decoded_size;
1762         }
1763         if (ret_verity_data)
1764                 *ret_verity_data = TAKE_PTR(verity_filename);
1765         if (roothashsig_filename)
1766                 *ret_roothashsig = TAKE_PTR(roothashsig_filename);
1767
1768         return 1;
1769 }
1770
1771 int dissected_image_acquire_metadata(DissectedImage *m) {
1772
1773         enum {
1774                 META_HOSTNAME,
1775                 META_MACHINE_ID,
1776                 META_MACHINE_INFO,
1777                 META_OS_RELEASE,
1778                 _META_MAX,
1779         };
1780
1781         static const char *const paths[_META_MAX] = {
1782                 [META_HOSTNAME]     = "/etc/hostname\0",
1783                 [META_MACHINE_ID]   = "/etc/machine-id\0",
1784                 [META_MACHINE_INFO] = "/etc/machine-info\0",
1785                 [META_OS_RELEASE]   = "/etc/os-release\0"
1786                                       "/usr/lib/os-release\0",
1787         };
1788
1789         _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1790         _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
1791         _cleanup_(rmdir_and_freep) char *t = NULL;
1792         _cleanup_(sigkill_waitp) pid_t child = 0;
1793         sd_id128_t machine_id = SD_ID128_NULL;
1794         _cleanup_free_ char *hostname = NULL;
1795         unsigned n_meta_initialized = 0, k;
1796         int fds[2 * _META_MAX], r, v;
1797         ssize_t n;
1798
1799         BLOCK_SIGNALS(SIGCHLD);
1800
1801         assert(m);
1802
1803         for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
1804                 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
1805                         r = -errno;
1806                         goto finish;
1807                 }
1808
1809         r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
1810         if (r < 0)
1811                 goto finish;
1812
1813         if (pipe2(error_pipe, O_CLOEXEC) < 0) {
1814                 r = -errno;
1815                 goto finish;
1816         }
1817
1818         r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
1819         if (r < 0)
1820                 goto finish;
1821         if (r == 0) {
1822                 error_pipe[0] = safe_close(error_pipe[0]);
1823
1824                 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
1825                 if (r < 0) {
1826                         /* Let parent know the error */
1827                         (void) write(error_pipe[1], &r, sizeof(r));
1828
1829                         log_debug_errno(r, "Failed to mount dissected image: %m");
1830                         _exit(EXIT_FAILURE);
1831                 }
1832
1833                 for (k = 0; k < _META_MAX; k++) {
1834                         _cleanup_close_ int fd = -ENOENT;
1835                         const char *p;
1836
1837                         fds[2*k] = safe_close(fds[2*k]);
1838
1839                         NULSTR_FOREACH(p, paths[k]) {
1840                                 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
1841                                 if (fd >= 0)
1842                                         break;
1843                         }
1844                         if (fd < 0) {
1845                                 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
1846                                 fds[2*k+1] = safe_close(fds[2*k+1]);
1847                                 continue;
1848                         }
1849
1850                         r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
1851                         if (r < 0) {
1852                                 (void) write(error_pipe[1], &r, sizeof(r));
1853                                 _exit(EXIT_FAILURE);
1854                         }
1855
1856                         fds[2*k+1] = safe_close(fds[2*k+1]);
1857                 }
1858
1859                 _exit(EXIT_SUCCESS);
1860         }
1861
1862         error_pipe[1] = safe_close(error_pipe[1]);
1863
1864         for (k = 0; k < _META_MAX; k++) {
1865                 _cleanup_fclose_ FILE *f = NULL;
1866
1867                 fds[2*k+1] = safe_close(fds[2*k+1]);
1868
1869                 f = take_fdopen(&fds[2*k], "r");
1870                 if (!f) {
1871                         r = -errno;
1872                         goto finish;
1873                 }
1874
1875                 switch (k) {
1876
1877                 case META_HOSTNAME:
1878                         r = read_etc_hostname_stream(f, &hostname);
1879                         if (r < 0)
1880                                 log_debug_errno(r, "Failed to read /etc/hostname: %m");
1881
1882                         break;
1883
1884                 case META_MACHINE_ID: {
1885                         _cleanup_free_ char *line = NULL;
1886
1887                         r = read_line(f, LONG_LINE_MAX, &line);
1888                         if (r < 0)
1889                                 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
1890                         else if (r == 33) {
1891                                 r = sd_id128_from_string(line, &machine_id);
1892                                 if (r < 0)
1893                                         log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
1894                         } else if (r == 0)
1895                                 log_debug("/etc/machine-id file is empty.");
1896                         else
1897                                 log_debug("/etc/machine-id has unexpected length %i.", r);
1898
1899                         break;
1900                 }
1901
1902                 case META_MACHINE_INFO:
1903                         r = load_env_file_pairs(f, "machine-info", &machine_info);
1904                         if (r < 0)
1905                                 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
1906
1907                         break;
1908
1909                 case META_OS_RELEASE:
1910                         r = load_env_file_pairs(f, "os-release", &os_release);
1911                         if (r < 0)
1912                                 log_debug_errno(r, "Failed to read OS release file: %m");
1913
1914                         break;
1915                 }
1916         }
1917
1918         r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
1919         child = 0;
1920         if (r < 0)
1921                 return r;
1922
1923         n = read(error_pipe[0], &v, sizeof(v));
1924         if (n < 0)
1925                 return -errno;
1926         if (n == sizeof(v))
1927                 return v; /* propagate error sent to us from child */
1928         if (n != 0)
1929                 return -EIO;
1930
1931         if (r != EXIT_SUCCESS)
1932                 return -EPROTO;
1933
1934         free_and_replace(m->hostname, hostname);
1935         m->machine_id = machine_id;
1936         strv_free_and_replace(m->machine_info, machine_info);
1937         strv_free_and_replace(m->os_release, os_release);
1938
1939 finish:
1940         for (k = 0; k < n_meta_initialized; k++)
1941                 safe_close_pair(fds + 2*k);
1942
1943         return r;
1944 }
1945
1946 int dissect_image_and_warn(
1947                 int fd,
1948                 const char *name,
1949                 const void *root_hash,
1950                 size_t root_hash_size,
1951                 const char *verity_data,
1952                 const MountOptions *mount_options,
1953                 DissectImageFlags flags,
1954                 DissectedImage **ret) {
1955
1956         _cleanup_free_ char *buffer = NULL;
1957         int r;
1958
1959         if (!name) {
1960                 r = fd_get_path(fd, &buffer);
1961                 if (r < 0)
1962                         return r;
1963
1964                 name = buffer;
1965         }
1966
1967         r = dissect_image(fd, root_hash, root_hash_size, verity_data, mount_options, flags, ret);
1968
1969         switch (r) {
1970
1971         case -EOPNOTSUPP:
1972                 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
1973
1974         case -ENOPKG:
1975                 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
1976
1977         case -EADDRNOTAVAIL:
1978                 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
1979
1980         case -ENOTUNIQ:
1981                 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
1982
1983         case -ENXIO:
1984                 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
1985
1986         case -EPROTONOSUPPORT:
1987                 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
1988
1989         default:
1990                 if (r < 0)
1991                         return log_error_errno(r, "Failed to dissect image '%s': %m", name);
1992
1993                 return r;
1994         }
1995 }
1996
1997 bool dissected_image_can_do_verity(const DissectedImage *image, unsigned partition_designator) {
1998         if (image->single_file_system)
1999                 return partition_designator == PARTITION_ROOT && image->can_verity;
2000
2001         return PARTITION_VERITY_OF(partition_designator) >= 0;
2002 }
2003
2004 bool dissected_image_has_verity(const DissectedImage *image, unsigned partition_designator) {
2005         int k;
2006
2007         if (image->single_file_system)
2008                 return partition_designator == PARTITION_ROOT && image->verity;
2009
2010         k = PARTITION_VERITY_OF(partition_designator);
2011         return k >= 0 && image->partitions[k].found;
2012 }
2013
2014 MountOptions* mount_options_free_all(MountOptions *options) {
2015         MountOptions *m;
2016
2017         while ((m = options)) {
2018                 LIST_REMOVE(mount_options, options, m);
2019                 free(m->options);
2020                 free(m);
2021         }
2022
2023         return NULL;
2024 }
2025
2026 const char* mount_options_from_part(const MountOptions *options, unsigned int partition_number) {
2027         MountOptions *m;
2028
2029         LIST_FOREACH(mount_options, m, (MountOptions *)options)
2030                 if (partition_number == m->partition_number && !isempty(m->options))
2031                         return m->options;
2032
2033         return NULL;
2034 }
2035
2036 int mount_image_privately_interactively(
2037                 const char *image,
2038                 DissectImageFlags flags,
2039                 char **ret_directory,
2040                 LoopDevice **ret_loop_device,
2041                 DecryptedImage **ret_decrypted_image) {
2042
2043         _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2044         _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2045         _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2046         _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2047         _cleanup_free_ char *temp = NULL;
2048         int r;
2049
2050         /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2051          * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2052          * easily. */
2053
2054         assert(image);
2055         assert(ret_directory);
2056         assert(ret_loop_device);
2057         assert(ret_decrypted_image);
2058
2059         r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2060         if (r < 0)
2061                 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2062
2063         r = loop_device_make_by_path(
2064                         image,
2065                         FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : O_RDWR,
2066                         FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
2067                         &d);
2068         if (r < 0)
2069                 return log_error_errno(r, "Failed to set up loopback device: %m");
2070
2071         r = dissect_image_and_warn(d->fd, image, NULL, 0, NULL, NULL, flags, &dissected_image);
2072         if (r < 0)
2073                 return r;
2074
2075         r = dissected_image_decrypt_interactively(dissected_image, NULL, NULL, 0, NULL, NULL, NULL, 0, flags, &decrypted_image);
2076         if (r < 0)
2077                 return r;
2078
2079         r = detach_mount_namespace();
2080         if (r < 0)
2081                 return log_error_errno(r, "Failed to detach mount namespace: %m");
2082
2083         r = mkdir_p(temp, 0700);
2084         if (r < 0)
2085                 return log_error_errno(r, "Failed to create mount point: %m");
2086
2087         created_dir = TAKE_PTR(temp);
2088
2089         r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, flags);
2090         if (r < 0)
2091                 return r;
2092
2093         if (decrypted_image) {
2094                 r = decrypted_image_relinquish(decrypted_image);
2095                 if (r < 0)
2096                         return log_error_errno(r, "Failed to relinquish DM devices: %m");
2097         }
2098
2099         loop_device_relinquish(d);
2100
2101         *ret_directory = TAKE_PTR(created_dir);
2102         *ret_loop_device = TAKE_PTR(d);
2103         *ret_decrypted_image = TAKE_PTR(decrypted_image);
2104
2105         return 0;
2106 }
2107
2108 static const char *const partition_designator_table[] = {
2109         [PARTITION_ROOT] = "root",
2110         [PARTITION_ROOT_SECONDARY] = "root-secondary",
2111         [PARTITION_HOME] = "home",
2112         [PARTITION_SRV] = "srv",
2113         [PARTITION_ESP] = "esp",
2114         [PARTITION_XBOOTLDR] = "xbootldr",
2115         [PARTITION_SWAP] = "swap",
2116         [PARTITION_ROOT_VERITY] = "root-verity",
2117         [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2118         [PARTITION_TMP] = "tmp",
2119         [PARTITION_VAR] = "var",
2120 };
2121
2122 DEFINE_STRING_TABLE_LOOKUP(partition_designator, int);